NSD

# NSD_complete This dataset repository stores a packaged release of `NSD_complete`. ## Why archives are used The original local dataset contains more than one million files, which exceeds the practical upload constraints of the dataset hub. To make upload and download stable, the dataset is published as top-level `.tar` archives that preserve the original folder structure. ## Archive layout - `COCO_IDs.tar` - `COCO_captions_recapted_Qw2VL.tar` - `NSD_fMRI_MNI_multi.tar` - `NSD_fMRI_MNI_single.tar` - `NSD_features.tar` - `NSD_imgs.tar` - `Visual_instruct_tuning_data.tar` - `nsddata.tar` - `short_COCO_caption.tar` - `root_files.tar` `root_files.tar` contains the loose files originally placed at the root of `NSD_complete`, including: - `COCO_73k_annots_curated.npy` - `brain2caption_qwen.json` - `raw_COCO.json` - `raw_COCO_with_idx.json` - `val_stim_multi_trial_data.npy` - `数据集说明.txt` ## Extraction Create a target directory and extract all archives into it: ```bash mkdir -p NSD_complete for f in *.tar; do tar -xf "$f" -C NSD_complete done ``` After extraction, the reconstructed directory layout matches the original `NSD_complete` dataset structure. ## Notes - Archives are uncompressed `.tar` files to reduce CPU overhead during packaging and upload. - Users can download only the sub-archives they need. - Please follow the original dataset and derivative data usage constraints when redistributing or using these files.